This adds a new feature to the Windows Terminal: "Workspaces"
Workspaces are very shamelessly inspired by Edge workspaces of the same
name.
The core idea is that when users name a window and they close that
window, we will persist that Windows layout and buffers, seperately from
the rest of window restoration. So a user can open a named window, open
some profiles, some panes, do some stuff in it, then close it, and we
will keep that state around for the next time the user opens that window
name.
Unnamed windows still behave the same. If you close an unnamed window,
and it's not the last window, then we won't persist the state of it.
To facilitate restoring named windows, we add a `openWorkspace` action.
This allows us to persist the open workspace action in the window layout
restoration path. So when we deserialize the list of tab layouts, and
open workspace action will tell us, hey, go retrieve this known
workspace from the state.json, instead of trying to serialize the window
state in two places.
As demoed in the video, we add a flyout to list the windows that the
user has open, and the named workspaces that they have saved. This
allows users to quickly reopen previously closed workspaces, as well as
quickly rename a window, thereby adding it to the list of saved
workspaces. This button can also be hidden using the theme settings.
Closes#17084
(this is a reboot of #20162, but with the event from #20311)
This adds a new feature to the Windows Terminal: "Workspaces"
Workspaces are very shamelessly inspired by Edge workspaces of the same
name.
The core idea is that when users name a window and they close that
window, we will persist that Windows layout and buffers, seperately from
the rest of window restoration. So a user can open a named window, open
some profiles, some panes, do some stuff in it, then close it, and we
will keep that state around for the next time the user opens that window
name.
Unnamed windows still behave the same. If you close an unnamed window,
and it's not the last window, then we won't persist the state of it.
To facilitate restoring named windows, we add a `openWorkspace` action.
This allows us to persist the open workspace action in the window layout
restoration path. So when we deserialize the list of tab layouts, and
open workspace action will tell us, hey, go retrieve this known
workspace from the state.json, instead of trying to serialize the window
state in two places.
As demoed in the video, we add a flyout to list the windows that the
user has open, and the named workspaces that they have saved. This
allows users to quickly reopen previously closed workspaces, as well as
quickly rename a window, thereby adding it to the list of saved
workspaces. This button can also be hidden using the theme settings.
Closes#17084
The idea with IControlSettings (and friends) was always that a consumer
of the terminal control could implement it in whatever way they pleased.
Windows Terminal (the application) was intended to be only one
consumer. It has a whole JSON settings model. Nobody wants to think
about JSON at the Terminal Control level. We could have an "adapter" in
TerminalApp, which spoke Terminal JSON Settings on one side and Terminal
Control on the other side.
That worked until we added the settings editor. The settings editor
needed to display a control, and that control's settings needed to be
based on the JSON settings. Oops. We took the expedient route of moving
the adapter into TerminalSettingsModel itself, and poking a bunch of
holes in it so that TerminalApp and TerminalSettingsEditor could tweak
it as needed.
Later, we doubled down on the control settings interface by having every
Terminal Control _make its own ControlSettings_ when we were going to do
the multi-process model. This reduced the number of IPC round trips for
every settings query to 0. Later we built color scheme previewing on top
of that--adding structs to carry color schemes and stuff which was
already in the Appearance config. Sheesh. Layers and layers and layers.
This pull request moves it back into its own library and strips it from
the surface of TerminalSettingsModel. It also deletes `ControlSettings`
and `struct CoreScheme`. That library is called
`TerminalSettingsAppAdapterLib`, and it contains a hidden WinRT
_implements_ type rather than a full-fledged activatable `runtimeclass`.
It also implements one-level inheritance on its own rather than using
IInheritable.
It adheres to the following principles:
- The control will never modify its settings in a way that is visible to
the control's consumer; therefore, none of the properties have setters
- The settings should never contain things of interest only to the
Application that the Application uses to communicate data _back to
itself_ (see `ProfileName`, removed in 68b723c and `KeyBindings`,
removed in fa09141). This generalizes to "we should never store stuff
in an unrelated object passed between layers solely for the purpose of
getting it back".
I made a few changes to the settings interface, including introducing a
new `ICoreScheme` interface that _only_ contains color scheme info. This
is designed to support the Preview/Set color scheme actions, which no
longer work by _app backing up the scheme and restoring it later._ All
of that machinery lives inside TermControl/ControlCore now.
`ICoreScheme` no longer supports `GetColorAtIndex`; you must read all 16
colors at the same time. I am not sorry. Every consumer did that
already, so now we have 15 fewer COM calls for every color scheme.
The new TerminalSettings is mostly consumed via
`com_ptr<TerminalSettings>`, so a bunch of `.` (projected) accesses had
to turn into `->` (com_ptr dereferencing) accesses.
I also realized, in the course of this work, that the old
TerminalSettings contained a partial hand-written reimplementation of
_every setting_ in `ControlProperties`. Every contributor had to add
every new setting to both places--why? I can't figure it out. I'm using
ControlProperties comprehensively now. I propagated any setting whose
default value was different from that in ControlProperties back to
ControlProperties.
This is part X in a series of pull requests that will remove all mention
of Microsoft.Terminal.Control and Microsoft.Terminal.Core from the
settings model. Once that is done, the settings model can consume _only_
the base WinRT types and build very early and test more easily.
Previewing is fun. I introduced a new place to stash an entire color
table on ControlCore, which we use to save the "active" colors while we
temporarily overwrite them. SetColorScheme is _also_ fun. We now have a
slot for overriding only the focused color scheme on ControlCore. It's
fine. It's clearer than "back up the focused appearance, overwrite the
focused appearance, create a child of the user's settings and apply the
color scheme to it, etc.".
There is a bug/design choice in color scheme overriding, which may or
may not matter: overlaying a color scheme on a terminal with an
unfocused appearance which _does not_ have its own color scheme will
result in the previously-deleted overridden focused color scheme peeking
through when the terminal is not focused.
I also got rid of our only in-product use of
`Terminal::CreateFromSettings` which required us to set `InitialRows`
and `InitialCols` on the incoming settings object (see core tenet 2).
Refs #19261
Refs #19314
Refs #19254
This pull request broadly rewrites how we handle all media resources in
the Terminal settings model.
## What is a media resource?
A media resource is any JSON property that refers to a file on disk,
including:
- `icon` on profile
- `backgroundImage` on profile (appearance)
- `pixelShaderPath` and `pixelShaderImagePath` on profile (appearance)
- `icon` on command and the new tab menu entries
The last two bullet points were newly discovered during the course of
this work.
## Description of Changes
In every place the settings model used to store a string for a media
path, it now stores an `IMediaResource`.
A media resource must be _resolved_ before it's used. When resolved, it
can report whether it is `Ok` (found, valid) and what the final
normalized path was.
This allows the settings model to apply some new behaviors.
One of those new behaviors is resolving media paths _relative to the
JSON file that referred to them._ This means fragments and user settings
can now contain _local_ images, pixel shaders and more and refer to them
by filename.
Relative path support requires us to track the path from which every
media resource "container" was read[^2]. For "big" objects like Profile,
we track it directly in the object and for each layer. This means that
fragments **updating** a profile pass their relative base path into the
mix. For some of the entries such as those in `newTabMenu`, we just wing
it (#19191). For everything that is recursively owned by a parent that
has a path (say each Command inside an ActionMap), we pass it in from
the parent during media resolution.
During resolution, we now track _exactly which layer_ an icon,
background image, or pixel shader path came from and read the "base
path" from only that layer. The base path is not inherited.
Another new behavior is in the handling of web and other URLs.
Canonical and a few other WSL distributors had to resort to web URLs for
icons because we did not support loading them from the package. Julia
tried to use `ms-appx://JuliaPackageNameHere/path/to/icon` for the same
reason. Neither was intended, and of the two the second _should_ have
worked but never could[^1].
For both `http(s?)` URLs and `ms-appx://` URLs which specify a package
name, we now strip everything except the filename. As an example...
If my fragment specifies `https://example.net/assets/foo.ico`, and my
fragment was loaded from `C:\Fragments`, Terminal will look *only* at
`C:\Fragments\foo.ico`.
This works today for Julia (they put their icon in the fragment folder
hoping that one day we would support this.) It will require some work
from existing WSL distributors.
I'm told that this is similar to how XML schema documents work.
Now, icons are special. They support _Emoji_ and _Segoe Icons_. This PR
adds an early pass to avoid resolving anything that looks like an
emoji.
This PR intentionally expands the heuristic definition of an emoji. It
used to only cover 1-2 code unit emoji, which prevented the use of any
emoji more complicated than "man in business suite levitating."
An icon path will now be considered an emoji or symbol icon if it is
composed of a single grapheme cluster (as measured by ICU.)
This is not perfect, as it errs on the side of allowing too many
things... but each of those things is technically a single grapheme
cluster and is a perfectly legal FontIcon ;)
Profile icons are _even more special_ than icons. They have an
additional fallback behavior which we had to preserve. When a profile
icon fails validation, or is expressly set to `null`, we fall back to
the EXE specified in the command line.
Because we do this fallback during resolution, _and the icon may be
inherited by any higher profile,_ we can only resolve it against the
commandline at the same level as the failed or nulled icon.
Therefore, if you specify `icon: null` in your `defaults` profile, it
will only ever resolve to `cmd.exe` for any profile that inherits it
(unless you change `defaults.commandline`).
This change expands support for the magic keywords `desktopWallpaper`
and `none` to all media paths (yes, even `pixelShaderPath`... but also,
`pixelShaderImagePath`!) It also expands support for _environment
variables_ to all of those places. Yes, we had like forty different
handlers for different types of string path. They are now uniform.
## Resource Validation
Media resources which are not found are "rejected". If a rejected
resource lives in _user_ settings, we will generate a warning and
display it.
In the future, we could detect this in the Settings UI and display a
warning inline.
## Surprises
I learned that `Windows.Foundation.Uri` parses file paths into `file://`
URIs, but does not offer you a way to get the original file path back
out. If you pass `C:\hello world`, _`Uri.Path`_ will return
`/C:/hello%20world`. I kid you not.
As a workaround, we bail out of URL handling if the `:` is too close to
the start (indicating an absolute file path).
## Testing
I added a narow test hook in the media resource resolver, which is
removed completely by link-time code generation. It is a real joy.
The test cases are all new and hopefully comprehensive.
Closes#19075Closes#16295Closes#10359 (except it doesn't support fonts)
Supersedes #16949 somewhat (`WT_SETTINGS_DIR`)
Refs #18679
Refs #19215 (future work)
Refs #19201 (future work)
Refs #19191 (future work)
[^1]: Handling a `ms-appx` path requires us to _add their package to our
dependency graph_ for the entire duration during which the resource will
be used. For us, that could be any time (like opening the command
palette for the first time!)
[^2]: We don't bother tracking where the defaults came from, because we
control everything about them.
This pull request removes the following vendored open source, in favor
of getting it from vcpkg:
- CLI11 2.4
- jsoncpp 1.9
- fmt 7.1.3
- gsl 3.1 (not vendored, but submoduled--arguably worse!)
Now that Visual Studio 2022 includes a built-in workload for vcpkg, the
onboarding process is much smoother. Terminal should only require the
vcpkg workload.
I've added some build rules that detect vcpkg via VS and via the user's
environment before falling back to a location in the source tree. The CI
pipeline will fall back to installing and bootstrapping vcpkg in
dep/vcpkg if necessary.
Some OSS has not been (and will not be) migrated:
- wyhash: ours is included directly in til/hash
- pcg_random: we have a stripped down copy compared to vcpkg
- stb_rect: vcpkg only ships *all of STB*; ours is a stripped down copy
- chromium numerics: vcpkg does not ship Chromium, especially not this
tiny fraction of Chromium
- dynamic_bitset and libpopcnt: removing in #17510
- interval_tree: no vcpkg equivalent
To support the needs of the inbox Windows build, I've split up our vcpkg
manifest into dependencies for all projects and dependencies just for
Terminal. To support this, we now offer a `terminal` feature. The vcpkg
rules in `common.build.pre.props` are set up to turn it on, whereas the
build rules we eventually write for the OS will not be.
Most of the work is concentrated in `common.build.pre.props`.
More prerequisite work for Action IDs - turns out if we add the action
IDs to the actions defined in `defaults.json` the string ends up being
too large and the compiler complains about it. Use a `.rc` file for
`defaults.json` instead and also for `enableColorSelection.json` +
`userDefaults.json`.
This pull request removes the need for the SettingsModel tests to run in
a UAP harness and puts them into the standard CI rotation.
This required some changes to `Run-Tests.ps1` to ensure that the right
`te.exe` is selected for each test harness. It's a bit annoying, but for
things that depend on a `resources.pri`, that file must be in the same
directory as the EXE that is hosting the test. Not the DLL, mind you,
the EXE. In our case, that's `TE.ProcessHost.exe`
The bulk of the change is honestly namespace tidying.
Co-authored-by: Mike Griese <migrie@microsoft.com>
Co-authored-by: Leonard Hecker <lhecker@microsoft.com>
